Never force full duplex.

Metod Kozelj metod.kozelj@rzs-hm.si
Sat Apr 1 05:52:37 2000


Hello,

On Fri, 31 Mar 2000, Dan Stromberg wrote:

> >       Never force full duplex.
> >       Never.  No, not then.  No, it's not a secret go-faster flag.
> > 
> > There is only one place where forcing full duplex is needed, and that is
> > when you are talking to an old or misdesigned switch that doesn't have
> > autonegotiation, and you cannot change the switch configuration.  If you can
> > change the switch configuration, change it back to half duplex and save
> > yourself future headaches.  You probably don't see any significant
> > performance improvement when using full duplex on switches without working
> > autonegotiation anyway.
> 
> I'm inclined to trust you on this Don, but I'd appreciate it if you
> could tell us more about the reasoning that went into this conclusion.
> I'm hesitant to recommend things on authority; I much prefer to
> understand (at least) a little bit about what I'm recommending.

There's another consideration about FD vs. HD: if you have, say 4
machines, all 100base-TX FD, and 3 of them want to push data to the fourth
one full speed, the bandwidth to the fourth one is too narrow. Switch'
buffers will become full and further packets from the first 3 machines 
will get silently dropped. High-level protocol will have to do the
re-transmissions after timeouts.

If all of them were HD, switch would insert collisions and low-level
protocol would do re-transmissions, which is faster.

However, I still use FD if possible :-)

Regards,
  Metod

Metod Kozelj

mailto:Metod.Kozelj@rzs-hm.si            /\  Ne posiljajte mi smeti ker grizem!
http://www.rzs-hm.si/                   /  \  Don't spam me for I bite!
_______________________________________/    \__________________________________

---- perl -e 'print $i=pack(c5,(41*2),sqrt(7056),(unpack(c,H)-2),oct(115),10);'


-------------------------------------------------------------------
To unsubscribe send a message body containing "unsubscribe"
to linux-tulip-request@beowulf.org